[PATCH] D143201: DAG: Lower single infinity is.fpclass tests to fcmp

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D143201#4174260 <https://reviews.llvm.org/D143201#4174260>, @sepavloff wrote:

> What is the advantage of such lowering?
>
> Implementation of `is_fpclass` as `fcmp` requires materialization of infinity constant. In contrast to zero constant, which often can be easily materialized, infinity usually requires loading constant from memory. So additional memory access is required as compared to the default lowering.

This also avoids a register bank copy if the target has separate integer and FP banks


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D143201/new/

https://reviews.llvm.org/D143201